]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/fs/namespace.c | |
3 | * | |
4 | * (C) Copyright Al Viro 2000, 2001 | |
5 | * Released under GPL v2. | |
6 | * | |
7 | * Based on code from fs/super.c, copyright Linus Torvalds and others. | |
8 | * Heavily rewritten. | |
9 | */ | |
10 | ||
1da177e4 LT |
11 | #include <linux/syscalls.h> |
12 | #include <linux/slab.h> | |
13 | #include <linux/sched.h> | |
14 | #include <linux/smp_lock.h> | |
15 | #include <linux/init.h> | |
15a67dd8 | 16 | #include <linux/kernel.h> |
1da177e4 LT |
17 | #include <linux/quotaops.h> |
18 | #include <linux/acct.h> | |
16f7e0fe | 19 | #include <linux/capability.h> |
1da177e4 | 20 | #include <linux/module.h> |
f20a9ead | 21 | #include <linux/sysfs.h> |
1da177e4 LT |
22 | #include <linux/seq_file.h> |
23 | #include <linux/namespace.h> | |
24 | #include <linux/namei.h> | |
25 | #include <linux/security.h> | |
26 | #include <linux/mount.h> | |
27 | #include <asm/uaccess.h> | |
28 | #include <asm/unistd.h> | |
07b20889 | 29 | #include "pnode.h" |
1da177e4 LT |
30 | |
31 | extern int __init init_rootfs(void); | |
32 | ||
1da177e4 | 33 | /* spinlock for vfsmount related operations, inplace of dcache_lock */ |
5addc5dd AV |
34 | __cacheline_aligned_in_smp DEFINE_SPINLOCK(vfsmount_lock); |
35 | ||
36 | static int event; | |
1da177e4 | 37 | |
fa3536cc | 38 | static struct list_head *mount_hashtable __read_mostly; |
6c231b7b | 39 | static int hash_mask __read_mostly, hash_bits __read_mostly; |
fa3536cc | 40 | static kmem_cache_t *mnt_cache __read_mostly; |
390c6843 | 41 | static struct rw_semaphore namespace_sem; |
1da177e4 | 42 | |
f87fd4c2 MS |
43 | /* /sys/fs */ |
44 | decl_subsys(fs, NULL, NULL); | |
45 | EXPORT_SYMBOL_GPL(fs_subsys); | |
46 | ||
1da177e4 LT |
47 | static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry) |
48 | { | |
b58fed8b RP |
49 | unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES); |
50 | tmp += ((unsigned long)dentry / L1_CACHE_BYTES); | |
1da177e4 LT |
51 | tmp = tmp + (tmp >> hash_bits); |
52 | return tmp & hash_mask; | |
53 | } | |
54 | ||
55 | struct vfsmount *alloc_vfsmnt(const char *name) | |
56 | { | |
b58fed8b | 57 | struct vfsmount *mnt = kmem_cache_alloc(mnt_cache, GFP_KERNEL); |
1da177e4 LT |
58 | if (mnt) { |
59 | memset(mnt, 0, sizeof(struct vfsmount)); | |
b58fed8b | 60 | atomic_set(&mnt->mnt_count, 1); |
1da177e4 LT |
61 | INIT_LIST_HEAD(&mnt->mnt_hash); |
62 | INIT_LIST_HEAD(&mnt->mnt_child); | |
63 | INIT_LIST_HEAD(&mnt->mnt_mounts); | |
64 | INIT_LIST_HEAD(&mnt->mnt_list); | |
55e700b9 | 65 | INIT_LIST_HEAD(&mnt->mnt_expire); |
03e06e68 | 66 | INIT_LIST_HEAD(&mnt->mnt_share); |
a58b0eb8 RP |
67 | INIT_LIST_HEAD(&mnt->mnt_slave_list); |
68 | INIT_LIST_HEAD(&mnt->mnt_slave); | |
1da177e4 | 69 | if (name) { |
b58fed8b | 70 | int size = strlen(name) + 1; |
1da177e4 LT |
71 | char *newname = kmalloc(size, GFP_KERNEL); |
72 | if (newname) { | |
73 | memcpy(newname, name, size); | |
74 | mnt->mnt_devname = newname; | |
75 | } | |
76 | } | |
77 | } | |
78 | return mnt; | |
79 | } | |
80 | ||
454e2398 DH |
81 | int simple_set_mnt(struct vfsmount *mnt, struct super_block *sb) |
82 | { | |
83 | mnt->mnt_sb = sb; | |
84 | mnt->mnt_root = dget(sb->s_root); | |
85 | return 0; | |
86 | } | |
87 | ||
88 | EXPORT_SYMBOL(simple_set_mnt); | |
89 | ||
1da177e4 LT |
90 | void free_vfsmnt(struct vfsmount *mnt) |
91 | { | |
92 | kfree(mnt->mnt_devname); | |
93 | kmem_cache_free(mnt_cache, mnt); | |
94 | } | |
95 | ||
96 | /* | |
a05964f3 RP |
97 | * find the first or last mount at @dentry on vfsmount @mnt depending on |
98 | * @dir. If @dir is set return the first mount else return the last mount. | |
1da177e4 | 99 | */ |
a05964f3 RP |
100 | struct vfsmount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry, |
101 | int dir) | |
1da177e4 | 102 | { |
b58fed8b RP |
103 | struct list_head *head = mount_hashtable + hash(mnt, dentry); |
104 | struct list_head *tmp = head; | |
1da177e4 LT |
105 | struct vfsmount *p, *found = NULL; |
106 | ||
1da177e4 | 107 | for (;;) { |
a05964f3 | 108 | tmp = dir ? tmp->next : tmp->prev; |
1da177e4 LT |
109 | p = NULL; |
110 | if (tmp == head) | |
111 | break; | |
112 | p = list_entry(tmp, struct vfsmount, mnt_hash); | |
113 | if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) { | |
a05964f3 | 114 | found = p; |
1da177e4 LT |
115 | break; |
116 | } | |
117 | } | |
1da177e4 LT |
118 | return found; |
119 | } | |
120 | ||
a05964f3 RP |
121 | /* |
122 | * lookup_mnt increments the ref count before returning | |
123 | * the vfsmount struct. | |
124 | */ | |
125 | struct vfsmount *lookup_mnt(struct vfsmount *mnt, struct dentry *dentry) | |
126 | { | |
127 | struct vfsmount *child_mnt; | |
128 | spin_lock(&vfsmount_lock); | |
129 | if ((child_mnt = __lookup_mnt(mnt, dentry, 1))) | |
130 | mntget(child_mnt); | |
131 | spin_unlock(&vfsmount_lock); | |
132 | return child_mnt; | |
133 | } | |
134 | ||
1da177e4 LT |
135 | static inline int check_mnt(struct vfsmount *mnt) |
136 | { | |
137 | return mnt->mnt_namespace == current->namespace; | |
138 | } | |
139 | ||
5addc5dd AV |
140 | static void touch_namespace(struct namespace *ns) |
141 | { | |
142 | if (ns) { | |
143 | ns->event = ++event; | |
144 | wake_up_interruptible(&ns->poll); | |
145 | } | |
146 | } | |
147 | ||
148 | static void __touch_namespace(struct namespace *ns) | |
149 | { | |
150 | if (ns && ns->event != event) { | |
151 | ns->event = event; | |
152 | wake_up_interruptible(&ns->poll); | |
153 | } | |
154 | } | |
155 | ||
1da177e4 LT |
156 | static void detach_mnt(struct vfsmount *mnt, struct nameidata *old_nd) |
157 | { | |
158 | old_nd->dentry = mnt->mnt_mountpoint; | |
159 | old_nd->mnt = mnt->mnt_parent; | |
160 | mnt->mnt_parent = mnt; | |
161 | mnt->mnt_mountpoint = mnt->mnt_root; | |
162 | list_del_init(&mnt->mnt_child); | |
163 | list_del_init(&mnt->mnt_hash); | |
164 | old_nd->dentry->d_mounted--; | |
165 | } | |
166 | ||
b90fa9ae RP |
167 | void mnt_set_mountpoint(struct vfsmount *mnt, struct dentry *dentry, |
168 | struct vfsmount *child_mnt) | |
169 | { | |
170 | child_mnt->mnt_parent = mntget(mnt); | |
171 | child_mnt->mnt_mountpoint = dget(dentry); | |
172 | dentry->d_mounted++; | |
173 | } | |
174 | ||
1da177e4 LT |
175 | static void attach_mnt(struct vfsmount *mnt, struct nameidata *nd) |
176 | { | |
b90fa9ae RP |
177 | mnt_set_mountpoint(nd->mnt, nd->dentry, mnt); |
178 | list_add_tail(&mnt->mnt_hash, mount_hashtable + | |
179 | hash(nd->mnt, nd->dentry)); | |
1da177e4 | 180 | list_add_tail(&mnt->mnt_child, &nd->mnt->mnt_mounts); |
b90fa9ae RP |
181 | } |
182 | ||
183 | /* | |
184 | * the caller must hold vfsmount_lock | |
185 | */ | |
186 | static void commit_tree(struct vfsmount *mnt) | |
187 | { | |
188 | struct vfsmount *parent = mnt->mnt_parent; | |
189 | struct vfsmount *m; | |
190 | LIST_HEAD(head); | |
191 | struct namespace *n = parent->mnt_namespace; | |
192 | ||
193 | BUG_ON(parent == mnt); | |
194 | ||
195 | list_add_tail(&head, &mnt->mnt_list); | |
196 | list_for_each_entry(m, &head, mnt_list) | |
197 | m->mnt_namespace = n; | |
198 | list_splice(&head, n->list.prev); | |
199 | ||
200 | list_add_tail(&mnt->mnt_hash, mount_hashtable + | |
201 | hash(parent, mnt->mnt_mountpoint)); | |
202 | list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); | |
203 | touch_namespace(n); | |
1da177e4 LT |
204 | } |
205 | ||
206 | static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root) | |
207 | { | |
208 | struct list_head *next = p->mnt_mounts.next; | |
209 | if (next == &p->mnt_mounts) { | |
210 | while (1) { | |
211 | if (p == root) | |
212 | return NULL; | |
213 | next = p->mnt_child.next; | |
214 | if (next != &p->mnt_parent->mnt_mounts) | |
215 | break; | |
216 | p = p->mnt_parent; | |
217 | } | |
218 | } | |
219 | return list_entry(next, struct vfsmount, mnt_child); | |
220 | } | |
221 | ||
9676f0c6 RP |
222 | static struct vfsmount *skip_mnt_tree(struct vfsmount *p) |
223 | { | |
224 | struct list_head *prev = p->mnt_mounts.prev; | |
225 | while (prev != &p->mnt_mounts) { | |
226 | p = list_entry(prev, struct vfsmount, mnt_child); | |
227 | prev = p->mnt_mounts.prev; | |
228 | } | |
229 | return p; | |
230 | } | |
231 | ||
36341f64 RP |
232 | static struct vfsmount *clone_mnt(struct vfsmount *old, struct dentry *root, |
233 | int flag) | |
1da177e4 LT |
234 | { |
235 | struct super_block *sb = old->mnt_sb; | |
236 | struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname); | |
237 | ||
238 | if (mnt) { | |
239 | mnt->mnt_flags = old->mnt_flags; | |
240 | atomic_inc(&sb->s_active); | |
241 | mnt->mnt_sb = sb; | |
242 | mnt->mnt_root = dget(root); | |
243 | mnt->mnt_mountpoint = mnt->mnt_root; | |
244 | mnt->mnt_parent = mnt; | |
b90fa9ae | 245 | |
5afe0022 RP |
246 | if (flag & CL_SLAVE) { |
247 | list_add(&mnt->mnt_slave, &old->mnt_slave_list); | |
248 | mnt->mnt_master = old; | |
249 | CLEAR_MNT_SHARED(mnt); | |
250 | } else { | |
251 | if ((flag & CL_PROPAGATION) || IS_MNT_SHARED(old)) | |
252 | list_add(&mnt->mnt_share, &old->mnt_share); | |
253 | if (IS_MNT_SLAVE(old)) | |
254 | list_add(&mnt->mnt_slave, &old->mnt_slave); | |
255 | mnt->mnt_master = old->mnt_master; | |
256 | } | |
b90fa9ae RP |
257 | if (flag & CL_MAKE_SHARED) |
258 | set_mnt_shared(mnt); | |
1da177e4 LT |
259 | |
260 | /* stick the duplicate mount on the same expiry list | |
261 | * as the original if that was on one */ | |
36341f64 RP |
262 | if (flag & CL_EXPIRE) { |
263 | spin_lock(&vfsmount_lock); | |
264 | if (!list_empty(&old->mnt_expire)) | |
265 | list_add(&mnt->mnt_expire, &old->mnt_expire); | |
266 | spin_unlock(&vfsmount_lock); | |
267 | } | |
1da177e4 LT |
268 | } |
269 | return mnt; | |
270 | } | |
271 | ||
7b7b1ace | 272 | static inline void __mntput(struct vfsmount *mnt) |
1da177e4 LT |
273 | { |
274 | struct super_block *sb = mnt->mnt_sb; | |
275 | dput(mnt->mnt_root); | |
276 | free_vfsmnt(mnt); | |
277 | deactivate_super(sb); | |
278 | } | |
279 | ||
7b7b1ace AV |
280 | void mntput_no_expire(struct vfsmount *mnt) |
281 | { | |
282 | repeat: | |
283 | if (atomic_dec_and_lock(&mnt->mnt_count, &vfsmount_lock)) { | |
284 | if (likely(!mnt->mnt_pinned)) { | |
285 | spin_unlock(&vfsmount_lock); | |
286 | __mntput(mnt); | |
287 | return; | |
288 | } | |
289 | atomic_add(mnt->mnt_pinned + 1, &mnt->mnt_count); | |
290 | mnt->mnt_pinned = 0; | |
291 | spin_unlock(&vfsmount_lock); | |
292 | acct_auto_close_mnt(mnt); | |
293 | security_sb_umount_close(mnt); | |
294 | goto repeat; | |
295 | } | |
296 | } | |
297 | ||
298 | EXPORT_SYMBOL(mntput_no_expire); | |
299 | ||
300 | void mnt_pin(struct vfsmount *mnt) | |
301 | { | |
302 | spin_lock(&vfsmount_lock); | |
303 | mnt->mnt_pinned++; | |
304 | spin_unlock(&vfsmount_lock); | |
305 | } | |
306 | ||
307 | EXPORT_SYMBOL(mnt_pin); | |
308 | ||
309 | void mnt_unpin(struct vfsmount *mnt) | |
310 | { | |
311 | spin_lock(&vfsmount_lock); | |
312 | if (mnt->mnt_pinned) { | |
313 | atomic_inc(&mnt->mnt_count); | |
314 | mnt->mnt_pinned--; | |
315 | } | |
316 | spin_unlock(&vfsmount_lock); | |
317 | } | |
318 | ||
319 | EXPORT_SYMBOL(mnt_unpin); | |
1da177e4 LT |
320 | |
321 | /* iterator */ | |
322 | static void *m_start(struct seq_file *m, loff_t *pos) | |
323 | { | |
324 | struct namespace *n = m->private; | |
325 | struct list_head *p; | |
326 | loff_t l = *pos; | |
327 | ||
390c6843 | 328 | down_read(&namespace_sem); |
1da177e4 LT |
329 | list_for_each(p, &n->list) |
330 | if (!l--) | |
331 | return list_entry(p, struct vfsmount, mnt_list); | |
332 | return NULL; | |
333 | } | |
334 | ||
335 | static void *m_next(struct seq_file *m, void *v, loff_t *pos) | |
336 | { | |
337 | struct namespace *n = m->private; | |
338 | struct list_head *p = ((struct vfsmount *)v)->mnt_list.next; | |
339 | (*pos)++; | |
b58fed8b | 340 | return p == &n->list ? NULL : list_entry(p, struct vfsmount, mnt_list); |
1da177e4 LT |
341 | } |
342 | ||
343 | static void m_stop(struct seq_file *m, void *v) | |
344 | { | |
390c6843 | 345 | up_read(&namespace_sem); |
1da177e4 LT |
346 | } |
347 | ||
348 | static inline void mangle(struct seq_file *m, const char *s) | |
349 | { | |
350 | seq_escape(m, s, " \t\n\\"); | |
351 | } | |
352 | ||
353 | static int show_vfsmnt(struct seq_file *m, void *v) | |
354 | { | |
355 | struct vfsmount *mnt = v; | |
356 | int err = 0; | |
357 | static struct proc_fs_info { | |
358 | int flag; | |
359 | char *str; | |
360 | } fs_info[] = { | |
361 | { MS_SYNCHRONOUS, ",sync" }, | |
362 | { MS_DIRSYNC, ",dirsync" }, | |
363 | { MS_MANDLOCK, ",mand" }, | |
1da177e4 LT |
364 | { 0, NULL } |
365 | }; | |
366 | static struct proc_fs_info mnt_info[] = { | |
367 | { MNT_NOSUID, ",nosuid" }, | |
368 | { MNT_NODEV, ",nodev" }, | |
369 | { MNT_NOEXEC, ",noexec" }, | |
fc33a7bb CH |
370 | { MNT_NOATIME, ",noatime" }, |
371 | { MNT_NODIRATIME, ",nodiratime" }, | |
1da177e4 LT |
372 | { 0, NULL } |
373 | }; | |
374 | struct proc_fs_info *fs_infop; | |
375 | ||
376 | mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none"); | |
377 | seq_putc(m, ' '); | |
378 | seq_path(m, mnt, mnt->mnt_root, " \t\n\\"); | |
379 | seq_putc(m, ' '); | |
380 | mangle(m, mnt->mnt_sb->s_type->name); | |
381 | seq_puts(m, mnt->mnt_sb->s_flags & MS_RDONLY ? " ro" : " rw"); | |
382 | for (fs_infop = fs_info; fs_infop->flag; fs_infop++) { | |
383 | if (mnt->mnt_sb->s_flags & fs_infop->flag) | |
384 | seq_puts(m, fs_infop->str); | |
385 | } | |
386 | for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) { | |
387 | if (mnt->mnt_flags & fs_infop->flag) | |
388 | seq_puts(m, fs_infop->str); | |
389 | } | |
390 | if (mnt->mnt_sb->s_op->show_options) | |
391 | err = mnt->mnt_sb->s_op->show_options(m, mnt); | |
392 | seq_puts(m, " 0 0\n"); | |
393 | return err; | |
394 | } | |
395 | ||
396 | struct seq_operations mounts_op = { | |
397 | .start = m_start, | |
398 | .next = m_next, | |
399 | .stop = m_stop, | |
400 | .show = show_vfsmnt | |
401 | }; | |
402 | ||
b4629fe2 CL |
403 | static int show_vfsstat(struct seq_file *m, void *v) |
404 | { | |
405 | struct vfsmount *mnt = v; | |
406 | int err = 0; | |
407 | ||
408 | /* device */ | |
409 | if (mnt->mnt_devname) { | |
410 | seq_puts(m, "device "); | |
411 | mangle(m, mnt->mnt_devname); | |
412 | } else | |
413 | seq_puts(m, "no device"); | |
414 | ||
415 | /* mount point */ | |
416 | seq_puts(m, " mounted on "); | |
417 | seq_path(m, mnt, mnt->mnt_root, " \t\n\\"); | |
418 | seq_putc(m, ' '); | |
419 | ||
420 | /* file system type */ | |
421 | seq_puts(m, "with fstype "); | |
422 | mangle(m, mnt->mnt_sb->s_type->name); | |
423 | ||
424 | /* optional statistics */ | |
425 | if (mnt->mnt_sb->s_op->show_stats) { | |
426 | seq_putc(m, ' '); | |
427 | err = mnt->mnt_sb->s_op->show_stats(m, mnt); | |
428 | } | |
429 | ||
430 | seq_putc(m, '\n'); | |
431 | return err; | |
432 | } | |
433 | ||
434 | struct seq_operations mountstats_op = { | |
435 | .start = m_start, | |
436 | .next = m_next, | |
437 | .stop = m_stop, | |
438 | .show = show_vfsstat, | |
439 | }; | |
440 | ||
1da177e4 LT |
441 | /** |
442 | * may_umount_tree - check if a mount tree is busy | |
443 | * @mnt: root of mount tree | |
444 | * | |
445 | * This is called to check if a tree of mounts has any | |
446 | * open files, pwds, chroots or sub mounts that are | |
447 | * busy. | |
448 | */ | |
449 | int may_umount_tree(struct vfsmount *mnt) | |
450 | { | |
36341f64 RP |
451 | int actual_refs = 0; |
452 | int minimum_refs = 0; | |
453 | struct vfsmount *p; | |
1da177e4 LT |
454 | |
455 | spin_lock(&vfsmount_lock); | |
36341f64 | 456 | for (p = mnt; p; p = next_mnt(p, mnt)) { |
1da177e4 LT |
457 | actual_refs += atomic_read(&p->mnt_count); |
458 | minimum_refs += 2; | |
1da177e4 LT |
459 | } |
460 | spin_unlock(&vfsmount_lock); | |
461 | ||
462 | if (actual_refs > minimum_refs) | |
e3474a8e | 463 | return 0; |
1da177e4 | 464 | |
e3474a8e | 465 | return 1; |
1da177e4 LT |
466 | } |
467 | ||
468 | EXPORT_SYMBOL(may_umount_tree); | |
469 | ||
470 | /** | |
471 | * may_umount - check if a mount point is busy | |
472 | * @mnt: root of mount | |
473 | * | |
474 | * This is called to check if a mount point has any | |
475 | * open files, pwds, chroots or sub mounts. If the | |
476 | * mount has sub mounts this will return busy | |
477 | * regardless of whether the sub mounts are busy. | |
478 | * | |
479 | * Doesn't take quota and stuff into account. IOW, in some cases it will | |
480 | * give false negatives. The main reason why it's here is that we need | |
481 | * a non-destructive way to look for easily umountable filesystems. | |
482 | */ | |
483 | int may_umount(struct vfsmount *mnt) | |
484 | { | |
e3474a8e | 485 | int ret = 1; |
a05964f3 RP |
486 | spin_lock(&vfsmount_lock); |
487 | if (propagate_mount_busy(mnt, 2)) | |
e3474a8e | 488 | ret = 0; |
a05964f3 RP |
489 | spin_unlock(&vfsmount_lock); |
490 | return ret; | |
1da177e4 LT |
491 | } |
492 | ||
493 | EXPORT_SYMBOL(may_umount); | |
494 | ||
b90fa9ae | 495 | void release_mounts(struct list_head *head) |
70fbcdf4 RP |
496 | { |
497 | struct vfsmount *mnt; | |
bf066c7d | 498 | while (!list_empty(head)) { |
70fbcdf4 RP |
499 | mnt = list_entry(head->next, struct vfsmount, mnt_hash); |
500 | list_del_init(&mnt->mnt_hash); | |
501 | if (mnt->mnt_parent != mnt) { | |
502 | struct dentry *dentry; | |
503 | struct vfsmount *m; | |
504 | spin_lock(&vfsmount_lock); | |
505 | dentry = mnt->mnt_mountpoint; | |
506 | m = mnt->mnt_parent; | |
507 | mnt->mnt_mountpoint = mnt->mnt_root; | |
508 | mnt->mnt_parent = mnt; | |
509 | spin_unlock(&vfsmount_lock); | |
510 | dput(dentry); | |
511 | mntput(m); | |
512 | } | |
513 | mntput(mnt); | |
514 | } | |
515 | } | |
516 | ||
a05964f3 | 517 | void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill) |
1da177e4 LT |
518 | { |
519 | struct vfsmount *p; | |
1da177e4 | 520 | |
1bfba4e8 AM |
521 | for (p = mnt; p; p = next_mnt(p, mnt)) |
522 | list_move(&p->mnt_hash, kill); | |
1da177e4 | 523 | |
a05964f3 RP |
524 | if (propagate) |
525 | propagate_umount(kill); | |
526 | ||
70fbcdf4 RP |
527 | list_for_each_entry(p, kill, mnt_hash) { |
528 | list_del_init(&p->mnt_expire); | |
529 | list_del_init(&p->mnt_list); | |
530 | __touch_namespace(p->mnt_namespace); | |
531 | p->mnt_namespace = NULL; | |
532 | list_del_init(&p->mnt_child); | |
533 | if (p->mnt_parent != p) | |
f30ac319 | 534 | p->mnt_mountpoint->d_mounted--; |
a05964f3 | 535 | change_mnt_propagation(p, MS_PRIVATE); |
1da177e4 LT |
536 | } |
537 | } | |
538 | ||
539 | static int do_umount(struct vfsmount *mnt, int flags) | |
540 | { | |
b58fed8b | 541 | struct super_block *sb = mnt->mnt_sb; |
1da177e4 | 542 | int retval; |
70fbcdf4 | 543 | LIST_HEAD(umount_list); |
1da177e4 LT |
544 | |
545 | retval = security_sb_umount(mnt, flags); | |
546 | if (retval) | |
547 | return retval; | |
548 | ||
549 | /* | |
550 | * Allow userspace to request a mountpoint be expired rather than | |
551 | * unmounting unconditionally. Unmount only happens if: | |
552 | * (1) the mark is already set (the mark is cleared by mntput()) | |
553 | * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] | |
554 | */ | |
555 | if (flags & MNT_EXPIRE) { | |
556 | if (mnt == current->fs->rootmnt || | |
557 | flags & (MNT_FORCE | MNT_DETACH)) | |
558 | return -EINVAL; | |
559 | ||
560 | if (atomic_read(&mnt->mnt_count) != 2) | |
561 | return -EBUSY; | |
562 | ||
563 | if (!xchg(&mnt->mnt_expiry_mark, 1)) | |
564 | return -EAGAIN; | |
565 | } | |
566 | ||
567 | /* | |
568 | * If we may have to abort operations to get out of this | |
569 | * mount, and they will themselves hold resources we must | |
570 | * allow the fs to do things. In the Unix tradition of | |
571 | * 'Gee thats tricky lets do it in userspace' the umount_begin | |
572 | * might fail to complete on the first run through as other tasks | |
573 | * must return, and the like. Thats for the mount program to worry | |
574 | * about for the moment. | |
575 | */ | |
576 | ||
577 | lock_kernel(); | |
8b512d9a TM |
578 | if (sb->s_op->umount_begin) |
579 | sb->s_op->umount_begin(mnt, flags); | |
1da177e4 LT |
580 | unlock_kernel(); |
581 | ||
582 | /* | |
583 | * No sense to grab the lock for this test, but test itself looks | |
584 | * somewhat bogus. Suggestions for better replacement? | |
585 | * Ho-hum... In principle, we might treat that as umount + switch | |
586 | * to rootfs. GC would eventually take care of the old vfsmount. | |
587 | * Actually it makes sense, especially if rootfs would contain a | |
588 | * /reboot - static binary that would close all descriptors and | |
589 | * call reboot(9). Then init(8) could umount root and exec /reboot. | |
590 | */ | |
591 | if (mnt == current->fs->rootmnt && !(flags & MNT_DETACH)) { | |
592 | /* | |
593 | * Special case for "unmounting" root ... | |
594 | * we just try to remount it readonly. | |
595 | */ | |
596 | down_write(&sb->s_umount); | |
597 | if (!(sb->s_flags & MS_RDONLY)) { | |
598 | lock_kernel(); | |
599 | DQUOT_OFF(sb); | |
600 | retval = do_remount_sb(sb, MS_RDONLY, NULL, 0); | |
601 | unlock_kernel(); | |
602 | } | |
603 | up_write(&sb->s_umount); | |
604 | return retval; | |
605 | } | |
606 | ||
390c6843 | 607 | down_write(&namespace_sem); |
1da177e4 | 608 | spin_lock(&vfsmount_lock); |
5addc5dd | 609 | event++; |
1da177e4 | 610 | |
1da177e4 | 611 | retval = -EBUSY; |
a05964f3 | 612 | if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) { |
1da177e4 | 613 | if (!list_empty(&mnt->mnt_list)) |
a05964f3 | 614 | umount_tree(mnt, 1, &umount_list); |
1da177e4 LT |
615 | retval = 0; |
616 | } | |
617 | spin_unlock(&vfsmount_lock); | |
618 | if (retval) | |
619 | security_sb_umount_busy(mnt); | |
390c6843 | 620 | up_write(&namespace_sem); |
70fbcdf4 | 621 | release_mounts(&umount_list); |
1da177e4 LT |
622 | return retval; |
623 | } | |
624 | ||
625 | /* | |
626 | * Now umount can handle mount points as well as block devices. | |
627 | * This is important for filesystems which use unnamed block devices. | |
628 | * | |
629 | * We now support a flag for forced unmount like the other 'big iron' | |
630 | * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD | |
631 | */ | |
632 | ||
633 | asmlinkage long sys_umount(char __user * name, int flags) | |
634 | { | |
635 | struct nameidata nd; | |
636 | int retval; | |
637 | ||
638 | retval = __user_walk(name, LOOKUP_FOLLOW, &nd); | |
639 | if (retval) | |
640 | goto out; | |
641 | retval = -EINVAL; | |
642 | if (nd.dentry != nd.mnt->mnt_root) | |
643 | goto dput_and_out; | |
644 | if (!check_mnt(nd.mnt)) | |
645 | goto dput_and_out; | |
646 | ||
647 | retval = -EPERM; | |
648 | if (!capable(CAP_SYS_ADMIN)) | |
649 | goto dput_and_out; | |
650 | ||
651 | retval = do_umount(nd.mnt, flags); | |
652 | dput_and_out: | |
653 | path_release_on_umount(&nd); | |
654 | out: | |
655 | return retval; | |
656 | } | |
657 | ||
658 | #ifdef __ARCH_WANT_SYS_OLDUMOUNT | |
659 | ||
660 | /* | |
b58fed8b | 661 | * The 2.0 compatible umount. No flags. |
1da177e4 | 662 | */ |
1da177e4 LT |
663 | asmlinkage long sys_oldumount(char __user * name) |
664 | { | |
b58fed8b | 665 | return sys_umount(name, 0); |
1da177e4 LT |
666 | } |
667 | ||
668 | #endif | |
669 | ||
670 | static int mount_is_safe(struct nameidata *nd) | |
671 | { | |
672 | if (capable(CAP_SYS_ADMIN)) | |
673 | return 0; | |
674 | return -EPERM; | |
675 | #ifdef notyet | |
676 | if (S_ISLNK(nd->dentry->d_inode->i_mode)) | |
677 | return -EPERM; | |
678 | if (nd->dentry->d_inode->i_mode & S_ISVTX) { | |
679 | if (current->uid != nd->dentry->d_inode->i_uid) | |
680 | return -EPERM; | |
681 | } | |
e4543edd | 682 | if (vfs_permission(nd, MAY_WRITE)) |
1da177e4 LT |
683 | return -EPERM; |
684 | return 0; | |
685 | #endif | |
686 | } | |
687 | ||
b58fed8b | 688 | static int lives_below_in_same_fs(struct dentry *d, struct dentry *dentry) |
1da177e4 LT |
689 | { |
690 | while (1) { | |
691 | if (d == dentry) | |
692 | return 1; | |
693 | if (d == NULL || d == d->d_parent) | |
694 | return 0; | |
695 | d = d->d_parent; | |
696 | } | |
697 | } | |
698 | ||
b90fa9ae | 699 | struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry, |
36341f64 | 700 | int flag) |
1da177e4 LT |
701 | { |
702 | struct vfsmount *res, *p, *q, *r, *s; | |
1da177e4 LT |
703 | struct nameidata nd; |
704 | ||
9676f0c6 RP |
705 | if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt)) |
706 | return NULL; | |
707 | ||
36341f64 | 708 | res = q = clone_mnt(mnt, dentry, flag); |
1da177e4 LT |
709 | if (!q) |
710 | goto Enomem; | |
711 | q->mnt_mountpoint = mnt->mnt_mountpoint; | |
712 | ||
713 | p = mnt; | |
fdadd65f | 714 | list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { |
1da177e4 LT |
715 | if (!lives_below_in_same_fs(r->mnt_mountpoint, dentry)) |
716 | continue; | |
717 | ||
718 | for (s = r; s; s = next_mnt(s, r)) { | |
9676f0c6 RP |
719 | if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) { |
720 | s = skip_mnt_tree(s); | |
721 | continue; | |
722 | } | |
1da177e4 LT |
723 | while (p != s->mnt_parent) { |
724 | p = p->mnt_parent; | |
725 | q = q->mnt_parent; | |
726 | } | |
727 | p = s; | |
728 | nd.mnt = q; | |
729 | nd.dentry = p->mnt_mountpoint; | |
36341f64 | 730 | q = clone_mnt(p, p->mnt_root, flag); |
1da177e4 LT |
731 | if (!q) |
732 | goto Enomem; | |
733 | spin_lock(&vfsmount_lock); | |
734 | list_add_tail(&q->mnt_list, &res->mnt_list); | |
735 | attach_mnt(q, &nd); | |
736 | spin_unlock(&vfsmount_lock); | |
737 | } | |
738 | } | |
739 | return res; | |
b58fed8b | 740 | Enomem: |
1da177e4 | 741 | if (res) { |
70fbcdf4 | 742 | LIST_HEAD(umount_list); |
1da177e4 | 743 | spin_lock(&vfsmount_lock); |
a05964f3 | 744 | umount_tree(res, 0, &umount_list); |
1da177e4 | 745 | spin_unlock(&vfsmount_lock); |
70fbcdf4 | 746 | release_mounts(&umount_list); |
1da177e4 LT |
747 | } |
748 | return NULL; | |
749 | } | |
750 | ||
b90fa9ae RP |
751 | /* |
752 | * @source_mnt : mount tree to be attached | |
21444403 RP |
753 | * @nd : place the mount tree @source_mnt is attached |
754 | * @parent_nd : if non-null, detach the source_mnt from its parent and | |
755 | * store the parent mount and mountpoint dentry. | |
756 | * (done when source_mnt is moved) | |
b90fa9ae RP |
757 | * |
758 | * NOTE: in the table below explains the semantics when a source mount | |
759 | * of a given type is attached to a destination mount of a given type. | |
9676f0c6 RP |
760 | * --------------------------------------------------------------------------- |
761 | * | BIND MOUNT OPERATION | | |
762 | * |************************************************************************** | |
763 | * | source-->| shared | private | slave | unbindable | | |
764 | * | dest | | | | | | |
765 | * | | | | | | | | |
766 | * | v | | | | | | |
767 | * |************************************************************************** | |
768 | * | shared | shared (++) | shared (+) | shared(+++)| invalid | | |
769 | * | | | | | | | |
770 | * |non-shared| shared (+) | private | slave (*) | invalid | | |
771 | * *************************************************************************** | |
b90fa9ae RP |
772 | * A bind operation clones the source mount and mounts the clone on the |
773 | * destination mount. | |
774 | * | |
775 | * (++) the cloned mount is propagated to all the mounts in the propagation | |
776 | * tree of the destination mount and the cloned mount is added to | |
777 | * the peer group of the source mount. | |
778 | * (+) the cloned mount is created under the destination mount and is marked | |
779 | * as shared. The cloned mount is added to the peer group of the source | |
780 | * mount. | |
5afe0022 RP |
781 | * (+++) the mount is propagated to all the mounts in the propagation tree |
782 | * of the destination mount and the cloned mount is made slave | |
783 | * of the same master as that of the source mount. The cloned mount | |
784 | * is marked as 'shared and slave'. | |
785 | * (*) the cloned mount is made a slave of the same master as that of the | |
786 | * source mount. | |
787 | * | |
9676f0c6 RP |
788 | * --------------------------------------------------------------------------- |
789 | * | MOVE MOUNT OPERATION | | |
790 | * |************************************************************************** | |
791 | * | source-->| shared | private | slave | unbindable | | |
792 | * | dest | | | | | | |
793 | * | | | | | | | | |
794 | * | v | | | | | | |
795 | * |************************************************************************** | |
796 | * | shared | shared (+) | shared (+) | shared(+++) | invalid | | |
797 | * | | | | | | | |
798 | * |non-shared| shared (+*) | private | slave (*) | unbindable | | |
799 | * *************************************************************************** | |
5afe0022 RP |
800 | * |
801 | * (+) the mount is moved to the destination. And is then propagated to | |
802 | * all the mounts in the propagation tree of the destination mount. | |
21444403 | 803 | * (+*) the mount is moved to the destination. |
5afe0022 RP |
804 | * (+++) the mount is moved to the destination and is then propagated to |
805 | * all the mounts belonging to the destination mount's propagation tree. | |
806 | * the mount is marked as 'shared and slave'. | |
807 | * (*) the mount continues to be a slave at the new location. | |
b90fa9ae RP |
808 | * |
809 | * if the source mount is a tree, the operations explained above is | |
810 | * applied to each mount in the tree. | |
811 | * Must be called without spinlocks held, since this function can sleep | |
812 | * in allocations. | |
813 | */ | |
814 | static int attach_recursive_mnt(struct vfsmount *source_mnt, | |
21444403 | 815 | struct nameidata *nd, struct nameidata *parent_nd) |
b90fa9ae RP |
816 | { |
817 | LIST_HEAD(tree_list); | |
818 | struct vfsmount *dest_mnt = nd->mnt; | |
819 | struct dentry *dest_dentry = nd->dentry; | |
820 | struct vfsmount *child, *p; | |
821 | ||
822 | if (propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list)) | |
823 | return -EINVAL; | |
824 | ||
825 | if (IS_MNT_SHARED(dest_mnt)) { | |
826 | for (p = source_mnt; p; p = next_mnt(p, source_mnt)) | |
827 | set_mnt_shared(p); | |
828 | } | |
829 | ||
830 | spin_lock(&vfsmount_lock); | |
21444403 RP |
831 | if (parent_nd) { |
832 | detach_mnt(source_mnt, parent_nd); | |
833 | attach_mnt(source_mnt, nd); | |
834 | touch_namespace(current->namespace); | |
835 | } else { | |
836 | mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt); | |
837 | commit_tree(source_mnt); | |
838 | } | |
b90fa9ae RP |
839 | |
840 | list_for_each_entry_safe(child, p, &tree_list, mnt_hash) { | |
841 | list_del_init(&child->mnt_hash); | |
842 | commit_tree(child); | |
843 | } | |
844 | spin_unlock(&vfsmount_lock); | |
845 | return 0; | |
846 | } | |
847 | ||
1da177e4 LT |
848 | static int graft_tree(struct vfsmount *mnt, struct nameidata *nd) |
849 | { | |
850 | int err; | |
851 | if (mnt->mnt_sb->s_flags & MS_NOUSER) | |
852 | return -EINVAL; | |
853 | ||
854 | if (S_ISDIR(nd->dentry->d_inode->i_mode) != | |
855 | S_ISDIR(mnt->mnt_root->d_inode->i_mode)) | |
856 | return -ENOTDIR; | |
857 | ||
858 | err = -ENOENT; | |
1b1dcc1b | 859 | mutex_lock(&nd->dentry->d_inode->i_mutex); |
1da177e4 LT |
860 | if (IS_DEADDIR(nd->dentry->d_inode)) |
861 | goto out_unlock; | |
862 | ||
863 | err = security_sb_check_sb(mnt, nd); | |
864 | if (err) | |
865 | goto out_unlock; | |
866 | ||
867 | err = -ENOENT; | |
b90fa9ae | 868 | if (IS_ROOT(nd->dentry) || !d_unhashed(nd->dentry)) |
21444403 | 869 | err = attach_recursive_mnt(mnt, nd, NULL); |
1da177e4 | 870 | out_unlock: |
1b1dcc1b | 871 | mutex_unlock(&nd->dentry->d_inode->i_mutex); |
1da177e4 LT |
872 | if (!err) |
873 | security_sb_post_addmount(mnt, nd); | |
874 | return err; | |
875 | } | |
876 | ||
07b20889 RP |
877 | /* |
878 | * recursively change the type of the mountpoint. | |
879 | */ | |
880 | static int do_change_type(struct nameidata *nd, int flag) | |
881 | { | |
882 | struct vfsmount *m, *mnt = nd->mnt; | |
883 | int recurse = flag & MS_REC; | |
884 | int type = flag & ~MS_REC; | |
885 | ||
886 | if (nd->dentry != nd->mnt->mnt_root) | |
887 | return -EINVAL; | |
888 | ||
889 | down_write(&namespace_sem); | |
890 | spin_lock(&vfsmount_lock); | |
891 | for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) | |
892 | change_mnt_propagation(m, type); | |
893 | spin_unlock(&vfsmount_lock); | |
894 | up_write(&namespace_sem); | |
895 | return 0; | |
896 | } | |
897 | ||
1da177e4 LT |
898 | /* |
899 | * do loopback mount. | |
900 | */ | |
eee391a6 | 901 | static int do_loopback(struct nameidata *nd, char *old_name, int recurse) |
1da177e4 LT |
902 | { |
903 | struct nameidata old_nd; | |
904 | struct vfsmount *mnt = NULL; | |
905 | int err = mount_is_safe(nd); | |
906 | if (err) | |
907 | return err; | |
908 | if (!old_name || !*old_name) | |
909 | return -EINVAL; | |
910 | err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd); | |
911 | if (err) | |
912 | return err; | |
913 | ||
390c6843 | 914 | down_write(&namespace_sem); |
1da177e4 | 915 | err = -EINVAL; |
9676f0c6 RP |
916 | if (IS_MNT_UNBINDABLE(old_nd.mnt)) |
917 | goto out; | |
918 | ||
ccd48bc7 AV |
919 | if (!check_mnt(nd->mnt) || !check_mnt(old_nd.mnt)) |
920 | goto out; | |
1da177e4 | 921 | |
ccd48bc7 AV |
922 | err = -ENOMEM; |
923 | if (recurse) | |
36341f64 | 924 | mnt = copy_tree(old_nd.mnt, old_nd.dentry, 0); |
ccd48bc7 | 925 | else |
36341f64 | 926 | mnt = clone_mnt(old_nd.mnt, old_nd.dentry, 0); |
ccd48bc7 AV |
927 | |
928 | if (!mnt) | |
929 | goto out; | |
930 | ||
ccd48bc7 AV |
931 | err = graft_tree(mnt, nd); |
932 | if (err) { | |
70fbcdf4 | 933 | LIST_HEAD(umount_list); |
1da177e4 | 934 | spin_lock(&vfsmount_lock); |
a05964f3 | 935 | umount_tree(mnt, 0, &umount_list); |
1da177e4 | 936 | spin_unlock(&vfsmount_lock); |
70fbcdf4 | 937 | release_mounts(&umount_list); |
5b83d2c5 | 938 | } |
1da177e4 | 939 | |
ccd48bc7 | 940 | out: |
390c6843 | 941 | up_write(&namespace_sem); |
1da177e4 LT |
942 | path_release(&old_nd); |
943 | return err; | |
944 | } | |
945 | ||
946 | /* | |
947 | * change filesystem flags. dir should be a physical root of filesystem. | |
948 | * If you've mounted a non-root directory somewhere and want to do remount | |
949 | * on it - tough luck. | |
950 | */ | |
1da177e4 LT |
951 | static int do_remount(struct nameidata *nd, int flags, int mnt_flags, |
952 | void *data) | |
953 | { | |
954 | int err; | |
b58fed8b | 955 | struct super_block *sb = nd->mnt->mnt_sb; |
1da177e4 LT |
956 | |
957 | if (!capable(CAP_SYS_ADMIN)) | |
958 | return -EPERM; | |
959 | ||
960 | if (!check_mnt(nd->mnt)) | |
961 | return -EINVAL; | |
962 | ||
963 | if (nd->dentry != nd->mnt->mnt_root) | |
964 | return -EINVAL; | |
965 | ||
966 | down_write(&sb->s_umount); | |
967 | err = do_remount_sb(sb, flags, data, 0); | |
968 | if (!err) | |
b58fed8b | 969 | nd->mnt->mnt_flags = mnt_flags; |
1da177e4 LT |
970 | up_write(&sb->s_umount); |
971 | if (!err) | |
972 | security_sb_post_remount(nd->mnt, flags, data); | |
973 | return err; | |
974 | } | |
975 | ||
9676f0c6 RP |
976 | static inline int tree_contains_unbindable(struct vfsmount *mnt) |
977 | { | |
978 | struct vfsmount *p; | |
979 | for (p = mnt; p; p = next_mnt(p, mnt)) { | |
980 | if (IS_MNT_UNBINDABLE(p)) | |
981 | return 1; | |
982 | } | |
983 | return 0; | |
984 | } | |
985 | ||
1da177e4 LT |
986 | static int do_move_mount(struct nameidata *nd, char *old_name) |
987 | { | |
988 | struct nameidata old_nd, parent_nd; | |
989 | struct vfsmount *p; | |
990 | int err = 0; | |
991 | if (!capable(CAP_SYS_ADMIN)) | |
992 | return -EPERM; | |
993 | if (!old_name || !*old_name) | |
994 | return -EINVAL; | |
995 | err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd); | |
996 | if (err) | |
997 | return err; | |
998 | ||
390c6843 | 999 | down_write(&namespace_sem); |
b58fed8b | 1000 | while (d_mountpoint(nd->dentry) && follow_down(&nd->mnt, &nd->dentry)) |
1da177e4 LT |
1001 | ; |
1002 | err = -EINVAL; | |
1003 | if (!check_mnt(nd->mnt) || !check_mnt(old_nd.mnt)) | |
1004 | goto out; | |
1005 | ||
1006 | err = -ENOENT; | |
1b1dcc1b | 1007 | mutex_lock(&nd->dentry->d_inode->i_mutex); |
1da177e4 LT |
1008 | if (IS_DEADDIR(nd->dentry->d_inode)) |
1009 | goto out1; | |
1010 | ||
1da177e4 | 1011 | if (!IS_ROOT(nd->dentry) && d_unhashed(nd->dentry)) |
21444403 | 1012 | goto out1; |
1da177e4 LT |
1013 | |
1014 | err = -EINVAL; | |
1015 | if (old_nd.dentry != old_nd.mnt->mnt_root) | |
21444403 | 1016 | goto out1; |
1da177e4 LT |
1017 | |
1018 | if (old_nd.mnt == old_nd.mnt->mnt_parent) | |
21444403 | 1019 | goto out1; |
1da177e4 LT |
1020 | |
1021 | if (S_ISDIR(nd->dentry->d_inode->i_mode) != | |
1022 | S_ISDIR(old_nd.dentry->d_inode->i_mode)) | |
21444403 RP |
1023 | goto out1; |
1024 | /* | |
1025 | * Don't move a mount residing in a shared parent. | |
1026 | */ | |
1027 | if (old_nd.mnt->mnt_parent && IS_MNT_SHARED(old_nd.mnt->mnt_parent)) | |
1028 | goto out1; | |
9676f0c6 RP |
1029 | /* |
1030 | * Don't move a mount tree containing unbindable mounts to a destination | |
1031 | * mount which is shared. | |
1032 | */ | |
1033 | if (IS_MNT_SHARED(nd->mnt) && tree_contains_unbindable(old_nd.mnt)) | |
1034 | goto out1; | |
1da177e4 | 1035 | err = -ELOOP; |
b58fed8b | 1036 | for (p = nd->mnt; p->mnt_parent != p; p = p->mnt_parent) |
1da177e4 | 1037 | if (p == old_nd.mnt) |
21444403 | 1038 | goto out1; |
1da177e4 | 1039 | |
21444403 RP |
1040 | if ((err = attach_recursive_mnt(old_nd.mnt, nd, &parent_nd))) |
1041 | goto out1; | |
1da177e4 | 1042 | |
21444403 | 1043 | spin_lock(&vfsmount_lock); |
1da177e4 LT |
1044 | /* if the mount is moved, it should no longer be expire |
1045 | * automatically */ | |
55e700b9 | 1046 | list_del_init(&old_nd.mnt->mnt_expire); |
1da177e4 LT |
1047 | spin_unlock(&vfsmount_lock); |
1048 | out1: | |
1b1dcc1b | 1049 | mutex_unlock(&nd->dentry->d_inode->i_mutex); |
1da177e4 | 1050 | out: |
390c6843 | 1051 | up_write(&namespace_sem); |
1da177e4 LT |
1052 | if (!err) |
1053 | path_release(&parent_nd); | |
1054 | path_release(&old_nd); | |
1055 | return err; | |
1056 | } | |
1057 | ||
1058 | /* | |
1059 | * create a new mount for userspace and request it to be added into the | |
1060 | * namespace's tree | |
1061 | */ | |
1062 | static int do_new_mount(struct nameidata *nd, char *type, int flags, | |
1063 | int mnt_flags, char *name, void *data) | |
1064 | { | |
1065 | struct vfsmount *mnt; | |
1066 | ||
1067 | if (!type || !memchr(type, 0, PAGE_SIZE)) | |
1068 | return -EINVAL; | |
1069 | ||
1070 | /* we need capabilities... */ | |
1071 | if (!capable(CAP_SYS_ADMIN)) | |
1072 | return -EPERM; | |
1073 | ||
1074 | mnt = do_kern_mount(type, flags, name, data); | |
1075 | if (IS_ERR(mnt)) | |
1076 | return PTR_ERR(mnt); | |
1077 | ||
1078 | return do_add_mount(mnt, nd, mnt_flags, NULL); | |
1079 | } | |
1080 | ||
1081 | /* | |
1082 | * add a mount into a namespace's mount tree | |
1083 | * - provide the option of adding the new mount to an expiration list | |
1084 | */ | |
1085 | int do_add_mount(struct vfsmount *newmnt, struct nameidata *nd, | |
1086 | int mnt_flags, struct list_head *fslist) | |
1087 | { | |
1088 | int err; | |
1089 | ||
390c6843 | 1090 | down_write(&namespace_sem); |
1da177e4 | 1091 | /* Something was mounted here while we slept */ |
b58fed8b | 1092 | while (d_mountpoint(nd->dentry) && follow_down(&nd->mnt, &nd->dentry)) |
1da177e4 LT |
1093 | ; |
1094 | err = -EINVAL; | |
1095 | if (!check_mnt(nd->mnt)) | |
1096 | goto unlock; | |
1097 | ||
1098 | /* Refuse the same filesystem on the same mount point */ | |
1099 | err = -EBUSY; | |
1100 | if (nd->mnt->mnt_sb == newmnt->mnt_sb && | |
1101 | nd->mnt->mnt_root == nd->dentry) | |
1102 | goto unlock; | |
1103 | ||
1104 | err = -EINVAL; | |
1105 | if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode)) | |
1106 | goto unlock; | |
1107 | ||
1108 | newmnt->mnt_flags = mnt_flags; | |
5b83d2c5 RP |
1109 | if ((err = graft_tree(newmnt, nd))) |
1110 | goto unlock; | |
1da177e4 | 1111 | |
5b83d2c5 | 1112 | if (fslist) { |
1da177e4 LT |
1113 | /* add to the specified expiration list */ |
1114 | spin_lock(&vfsmount_lock); | |
55e700b9 | 1115 | list_add_tail(&newmnt->mnt_expire, fslist); |
1da177e4 LT |
1116 | spin_unlock(&vfsmount_lock); |
1117 | } | |
390c6843 | 1118 | up_write(&namespace_sem); |
5b83d2c5 | 1119 | return 0; |
1da177e4 LT |
1120 | |
1121 | unlock: | |
390c6843 | 1122 | up_write(&namespace_sem); |
1da177e4 LT |
1123 | mntput(newmnt); |
1124 | return err; | |
1125 | } | |
1126 | ||
1127 | EXPORT_SYMBOL_GPL(do_add_mount); | |
1128 | ||
70fbcdf4 RP |
1129 | static void expire_mount(struct vfsmount *mnt, struct list_head *mounts, |
1130 | struct list_head *umounts) | |
24ca2af1 MS |
1131 | { |
1132 | spin_lock(&vfsmount_lock); | |
1133 | ||
ed42c879 MS |
1134 | /* |
1135 | * Check if mount is still attached, if not, let whoever holds it deal | |
1136 | * with the sucker | |
1137 | */ | |
1138 | if (mnt->mnt_parent == mnt) { | |
1139 | spin_unlock(&vfsmount_lock); | |
1140 | return; | |
1141 | } | |
1142 | ||
24ca2af1 MS |
1143 | /* |
1144 | * Check that it is still dead: the count should now be 2 - as | |
1145 | * contributed by the vfsmount parent and the mntget above | |
1146 | */ | |
a05964f3 | 1147 | if (!propagate_mount_busy(mnt, 2)) { |
24ca2af1 | 1148 | /* delete from the namespace */ |
5addc5dd | 1149 | touch_namespace(mnt->mnt_namespace); |
24ca2af1 | 1150 | list_del_init(&mnt->mnt_list); |
ac081153 | 1151 | mnt->mnt_namespace = NULL; |
a05964f3 | 1152 | umount_tree(mnt, 1, umounts); |
24ca2af1 | 1153 | spin_unlock(&vfsmount_lock); |
24ca2af1 MS |
1154 | } else { |
1155 | /* | |
1156 | * Someone brought it back to life whilst we didn't have any | |
1157 | * locks held so return it to the expiration list | |
1158 | */ | |
55e700b9 | 1159 | list_add_tail(&mnt->mnt_expire, mounts); |
24ca2af1 MS |
1160 | spin_unlock(&vfsmount_lock); |
1161 | } | |
1162 | } | |
1163 | ||
5528f911 TM |
1164 | /* |
1165 | * go through the vfsmounts we've just consigned to the graveyard to | |
1166 | * - check that they're still dead | |
1167 | * - delete the vfsmount from the appropriate namespace under lock | |
1168 | * - dispose of the corpse | |
1169 | */ | |
1170 | static void expire_mount_list(struct list_head *graveyard, struct list_head *mounts) | |
1171 | { | |
1172 | struct namespace *namespace; | |
1173 | struct vfsmount *mnt; | |
1174 | ||
1175 | while (!list_empty(graveyard)) { | |
1176 | LIST_HEAD(umounts); | |
1177 | mnt = list_entry(graveyard->next, struct vfsmount, mnt_expire); | |
1178 | list_del_init(&mnt->mnt_expire); | |
1179 | ||
1180 | /* don't do anything if the namespace is dead - all the | |
1181 | * vfsmounts from it are going away anyway */ | |
1182 | namespace = mnt->mnt_namespace; | |
1183 | if (!namespace || !namespace->root) | |
1184 | continue; | |
1185 | get_namespace(namespace); | |
1186 | ||
1187 | spin_unlock(&vfsmount_lock); | |
1188 | down_write(&namespace_sem); | |
1189 | expire_mount(mnt, mounts, &umounts); | |
1190 | up_write(&namespace_sem); | |
1191 | release_mounts(&umounts); | |
1192 | mntput(mnt); | |
1193 | put_namespace(namespace); | |
1194 | spin_lock(&vfsmount_lock); | |
1195 | } | |
1196 | } | |
1197 | ||
1da177e4 LT |
1198 | /* |
1199 | * process a list of expirable mountpoints with the intent of discarding any | |
1200 | * mountpoints that aren't in use and haven't been touched since last we came | |
1201 | * here | |
1202 | */ | |
1203 | void mark_mounts_for_expiry(struct list_head *mounts) | |
1204 | { | |
1da177e4 LT |
1205 | struct vfsmount *mnt, *next; |
1206 | LIST_HEAD(graveyard); | |
1207 | ||
1208 | if (list_empty(mounts)) | |
1209 | return; | |
1210 | ||
1211 | spin_lock(&vfsmount_lock); | |
1212 | ||
1213 | /* extract from the expiration list every vfsmount that matches the | |
1214 | * following criteria: | |
1215 | * - only referenced by its parent vfsmount | |
1216 | * - still marked for expiry (marked on the last call here; marks are | |
1217 | * cleared by mntput()) | |
1218 | */ | |
55e700b9 | 1219 | list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { |
1da177e4 LT |
1220 | if (!xchg(&mnt->mnt_expiry_mark, 1) || |
1221 | atomic_read(&mnt->mnt_count) != 1) | |
1222 | continue; | |
1223 | ||
1224 | mntget(mnt); | |
55e700b9 | 1225 | list_move(&mnt->mnt_expire, &graveyard); |
1da177e4 LT |
1226 | } |
1227 | ||
5528f911 | 1228 | expire_mount_list(&graveyard, mounts); |
1da177e4 | 1229 | |
5528f911 TM |
1230 | spin_unlock(&vfsmount_lock); |
1231 | } | |
1232 | ||
1233 | EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); | |
1234 | ||
1235 | /* | |
1236 | * Ripoff of 'select_parent()' | |
1237 | * | |
1238 | * search the list of submounts for a given mountpoint, and move any | |
1239 | * shrinkable submounts to the 'graveyard' list. | |
1240 | */ | |
1241 | static int select_submounts(struct vfsmount *parent, struct list_head *graveyard) | |
1242 | { | |
1243 | struct vfsmount *this_parent = parent; | |
1244 | struct list_head *next; | |
1245 | int found = 0; | |
1246 | ||
1247 | repeat: | |
1248 | next = this_parent->mnt_mounts.next; | |
1249 | resume: | |
1250 | while (next != &this_parent->mnt_mounts) { | |
1251 | struct list_head *tmp = next; | |
1252 | struct vfsmount *mnt = list_entry(tmp, struct vfsmount, mnt_child); | |
1253 | ||
1254 | next = tmp->next; | |
1255 | if (!(mnt->mnt_flags & MNT_SHRINKABLE)) | |
1da177e4 | 1256 | continue; |
5528f911 TM |
1257 | /* |
1258 | * Descend a level if the d_mounts list is non-empty. | |
1259 | */ | |
1260 | if (!list_empty(&mnt->mnt_mounts)) { | |
1261 | this_parent = mnt; | |
1262 | goto repeat; | |
1263 | } | |
1da177e4 | 1264 | |
5528f911 TM |
1265 | if (!propagate_mount_busy(mnt, 1)) { |
1266 | mntget(mnt); | |
1267 | list_move_tail(&mnt->mnt_expire, graveyard); | |
1268 | found++; | |
1269 | } | |
1da177e4 | 1270 | } |
5528f911 TM |
1271 | /* |
1272 | * All done at this level ... ascend and resume the search | |
1273 | */ | |
1274 | if (this_parent != parent) { | |
1275 | next = this_parent->mnt_child.next; | |
1276 | this_parent = this_parent->mnt_parent; | |
1277 | goto resume; | |
1278 | } | |
1279 | return found; | |
1280 | } | |
1281 | ||
1282 | /* | |
1283 | * process a list of expirable mountpoints with the intent of discarding any | |
1284 | * submounts of a specific parent mountpoint | |
1285 | */ | |
1286 | void shrink_submounts(struct vfsmount *mountpoint, struct list_head *mounts) | |
1287 | { | |
1288 | LIST_HEAD(graveyard); | |
1289 | int found; | |
1290 | ||
1291 | spin_lock(&vfsmount_lock); | |
1292 | ||
1293 | /* extract submounts of 'mountpoint' from the expiration list */ | |
1294 | while ((found = select_submounts(mountpoint, &graveyard)) != 0) | |
1295 | expire_mount_list(&graveyard, mounts); | |
1da177e4 LT |
1296 | |
1297 | spin_unlock(&vfsmount_lock); | |
1298 | } | |
1299 | ||
5528f911 | 1300 | EXPORT_SYMBOL_GPL(shrink_submounts); |
1da177e4 LT |
1301 | |
1302 | /* | |
1303 | * Some copy_from_user() implementations do not return the exact number of | |
1304 | * bytes remaining to copy on a fault. But copy_mount_options() requires that. | |
1305 | * Note that this function differs from copy_from_user() in that it will oops | |
1306 | * on bad values of `to', rather than returning a short copy. | |
1307 | */ | |
b58fed8b RP |
1308 | static long exact_copy_from_user(void *to, const void __user * from, |
1309 | unsigned long n) | |
1da177e4 LT |
1310 | { |
1311 | char *t = to; | |
1312 | const char __user *f = from; | |
1313 | char c; | |
1314 | ||
1315 | if (!access_ok(VERIFY_READ, from, n)) | |
1316 | return n; | |
1317 | ||
1318 | while (n) { | |
1319 | if (__get_user(c, f)) { | |
1320 | memset(t, 0, n); | |
1321 | break; | |
1322 | } | |
1323 | *t++ = c; | |
1324 | f++; | |
1325 | n--; | |
1326 | } | |
1327 | return n; | |
1328 | } | |
1329 | ||
b58fed8b | 1330 | int copy_mount_options(const void __user * data, unsigned long *where) |
1da177e4 LT |
1331 | { |
1332 | int i; | |
1333 | unsigned long page; | |
1334 | unsigned long size; | |
b58fed8b | 1335 | |
1da177e4 LT |
1336 | *where = 0; |
1337 | if (!data) | |
1338 | return 0; | |
1339 | ||
1340 | if (!(page = __get_free_page(GFP_KERNEL))) | |
1341 | return -ENOMEM; | |
1342 | ||
1343 | /* We only care that *some* data at the address the user | |
1344 | * gave us is valid. Just in case, we'll zero | |
1345 | * the remainder of the page. | |
1346 | */ | |
1347 | /* copy_from_user cannot cross TASK_SIZE ! */ | |
1348 | size = TASK_SIZE - (unsigned long)data; | |
1349 | if (size > PAGE_SIZE) | |
1350 | size = PAGE_SIZE; | |
1351 | ||
1352 | i = size - exact_copy_from_user((void *)page, data, size); | |
1353 | if (!i) { | |
b58fed8b | 1354 | free_page(page); |
1da177e4 LT |
1355 | return -EFAULT; |
1356 | } | |
1357 | if (i != PAGE_SIZE) | |
1358 | memset((char *)page + i, 0, PAGE_SIZE - i); | |
1359 | *where = page; | |
1360 | return 0; | |
1361 | } | |
1362 | ||
1363 | /* | |
1364 | * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to | |
1365 | * be given to the mount() call (ie: read-only, no-dev, no-suid etc). | |
1366 | * | |
1367 | * data is a (void *) that can point to any structure up to | |
1368 | * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent | |
1369 | * information (or be NULL). | |
1370 | * | |
1371 | * Pre-0.97 versions of mount() didn't have a flags word. | |
1372 | * When the flags word was introduced its top half was required | |
1373 | * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. | |
1374 | * Therefore, if this magic number is present, it carries no information | |
1375 | * and must be discarded. | |
1376 | */ | |
b58fed8b | 1377 | long do_mount(char *dev_name, char *dir_name, char *type_page, |
1da177e4 LT |
1378 | unsigned long flags, void *data_page) |
1379 | { | |
1380 | struct nameidata nd; | |
1381 | int retval = 0; | |
1382 | int mnt_flags = 0; | |
1383 | ||
1384 | /* Discard magic */ | |
1385 | if ((flags & MS_MGC_MSK) == MS_MGC_VAL) | |
1386 | flags &= ~MS_MGC_MSK; | |
1387 | ||
1388 | /* Basic sanity checks */ | |
1389 | ||
1390 | if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE)) | |
1391 | return -EINVAL; | |
1392 | if (dev_name && !memchr(dev_name, 0, PAGE_SIZE)) | |
1393 | return -EINVAL; | |
1394 | ||
1395 | if (data_page) | |
1396 | ((char *)data_page)[PAGE_SIZE - 1] = 0; | |
1397 | ||
1398 | /* Separate the per-mountpoint flags */ | |
1399 | if (flags & MS_NOSUID) | |
1400 | mnt_flags |= MNT_NOSUID; | |
1401 | if (flags & MS_NODEV) | |
1402 | mnt_flags |= MNT_NODEV; | |
1403 | if (flags & MS_NOEXEC) | |
1404 | mnt_flags |= MNT_NOEXEC; | |
fc33a7bb CH |
1405 | if (flags & MS_NOATIME) |
1406 | mnt_flags |= MNT_NOATIME; | |
1407 | if (flags & MS_NODIRATIME) | |
1408 | mnt_flags |= MNT_NODIRATIME; | |
1409 | ||
1410 | flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | | |
1411 | MS_NOATIME | MS_NODIRATIME); | |
1da177e4 LT |
1412 | |
1413 | /* ... and get the mountpoint */ | |
1414 | retval = path_lookup(dir_name, LOOKUP_FOLLOW, &nd); | |
1415 | if (retval) | |
1416 | return retval; | |
1417 | ||
1418 | retval = security_sb_mount(dev_name, &nd, type_page, flags, data_page); | |
1419 | if (retval) | |
1420 | goto dput_out; | |
1421 | ||
1422 | if (flags & MS_REMOUNT) | |
1423 | retval = do_remount(&nd, flags & ~MS_REMOUNT, mnt_flags, | |
1424 | data_page); | |
1425 | else if (flags & MS_BIND) | |
eee391a6 | 1426 | retval = do_loopback(&nd, dev_name, flags & MS_REC); |
9676f0c6 | 1427 | else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) |
07b20889 | 1428 | retval = do_change_type(&nd, flags); |
1da177e4 LT |
1429 | else if (flags & MS_MOVE) |
1430 | retval = do_move_mount(&nd, dev_name); | |
1431 | else | |
1432 | retval = do_new_mount(&nd, type_page, flags, mnt_flags, | |
1433 | dev_name, data_page); | |
1434 | dput_out: | |
1435 | path_release(&nd); | |
1436 | return retval; | |
1437 | } | |
1438 | ||
741a2951 JD |
1439 | /* |
1440 | * Allocate a new namespace structure and populate it with contents | |
1441 | * copied from the namespace of the passed in task structure. | |
1442 | */ | |
1443 | struct namespace *dup_namespace(struct task_struct *tsk, struct fs_struct *fs) | |
1da177e4 LT |
1444 | { |
1445 | struct namespace *namespace = tsk->namespace; | |
1446 | struct namespace *new_ns; | |
1447 | struct vfsmount *rootmnt = NULL, *pwdmnt = NULL, *altrootmnt = NULL; | |
1da177e4 LT |
1448 | struct vfsmount *p, *q; |
1449 | ||
1da177e4 LT |
1450 | new_ns = kmalloc(sizeof(struct namespace), GFP_KERNEL); |
1451 | if (!new_ns) | |
f13b8358 | 1452 | return NULL; |
1da177e4 LT |
1453 | |
1454 | atomic_set(&new_ns->count, 1); | |
1da177e4 | 1455 | INIT_LIST_HEAD(&new_ns->list); |
5addc5dd AV |
1456 | init_waitqueue_head(&new_ns->poll); |
1457 | new_ns->event = 0; | |
1da177e4 | 1458 | |
390c6843 | 1459 | down_write(&namespace_sem); |
1da177e4 | 1460 | /* First pass: copy the tree topology */ |
36341f64 | 1461 | new_ns->root = copy_tree(namespace->root, namespace->root->mnt_root, |
9676f0c6 | 1462 | CL_COPY_ALL | CL_EXPIRE); |
1da177e4 | 1463 | if (!new_ns->root) { |
390c6843 | 1464 | up_write(&namespace_sem); |
1da177e4 | 1465 | kfree(new_ns); |
f13b8358 | 1466 | return NULL; |
1da177e4 LT |
1467 | } |
1468 | spin_lock(&vfsmount_lock); | |
1469 | list_add_tail(&new_ns->list, &new_ns->root->mnt_list); | |
1470 | spin_unlock(&vfsmount_lock); | |
1471 | ||
1472 | /* | |
1473 | * Second pass: switch the tsk->fs->* elements and mark new vfsmounts | |
1474 | * as belonging to new namespace. We have already acquired a private | |
1475 | * fs_struct, so tsk->fs->lock is not needed. | |
1476 | */ | |
1477 | p = namespace->root; | |
1478 | q = new_ns->root; | |
1479 | while (p) { | |
1480 | q->mnt_namespace = new_ns; | |
1481 | if (fs) { | |
1482 | if (p == fs->rootmnt) { | |
1483 | rootmnt = p; | |
1484 | fs->rootmnt = mntget(q); | |
1485 | } | |
1486 | if (p == fs->pwdmnt) { | |
1487 | pwdmnt = p; | |
1488 | fs->pwdmnt = mntget(q); | |
1489 | } | |
1490 | if (p == fs->altrootmnt) { | |
1491 | altrootmnt = p; | |
1492 | fs->altrootmnt = mntget(q); | |
1493 | } | |
1494 | } | |
1495 | p = next_mnt(p, namespace->root); | |
1496 | q = next_mnt(q, new_ns->root); | |
1497 | } | |
390c6843 | 1498 | up_write(&namespace_sem); |
1da177e4 | 1499 | |
1da177e4 LT |
1500 | if (rootmnt) |
1501 | mntput(rootmnt); | |
1502 | if (pwdmnt) | |
1503 | mntput(pwdmnt); | |
1504 | if (altrootmnt) | |
1505 | mntput(altrootmnt); | |
1506 | ||
741a2951 JD |
1507 | return new_ns; |
1508 | } | |
1509 | ||
1510 | int copy_namespace(int flags, struct task_struct *tsk) | |
1511 | { | |
1512 | struct namespace *namespace = tsk->namespace; | |
1513 | struct namespace *new_ns; | |
1514 | int err = 0; | |
1515 | ||
1516 | if (!namespace) | |
1517 | return 0; | |
1518 | ||
1519 | get_namespace(namespace); | |
1520 | ||
1521 | if (!(flags & CLONE_NEWNS)) | |
1522 | return 0; | |
1523 | ||
1524 | if (!capable(CAP_SYS_ADMIN)) { | |
1525 | err = -EPERM; | |
1526 | goto out; | |
1527 | } | |
1528 | ||
1529 | new_ns = dup_namespace(tsk, tsk->fs); | |
1530 | if (!new_ns) { | |
1531 | err = -ENOMEM; | |
1532 | goto out; | |
1533 | } | |
1534 | ||
1535 | tsk->namespace = new_ns; | |
1da177e4 LT |
1536 | |
1537 | out: | |
1538 | put_namespace(namespace); | |
741a2951 | 1539 | return err; |
1da177e4 LT |
1540 | } |
1541 | ||
1542 | asmlinkage long sys_mount(char __user * dev_name, char __user * dir_name, | |
1543 | char __user * type, unsigned long flags, | |
1544 | void __user * data) | |
1545 | { | |
1546 | int retval; | |
1547 | unsigned long data_page; | |
1548 | unsigned long type_page; | |
1549 | unsigned long dev_page; | |
1550 | char *dir_page; | |
1551 | ||
b58fed8b | 1552 | retval = copy_mount_options(type, &type_page); |
1da177e4 LT |
1553 | if (retval < 0) |
1554 | return retval; | |
1555 | ||
1556 | dir_page = getname(dir_name); | |
1557 | retval = PTR_ERR(dir_page); | |
1558 | if (IS_ERR(dir_page)) | |
1559 | goto out1; | |
1560 | ||
b58fed8b | 1561 | retval = copy_mount_options(dev_name, &dev_page); |
1da177e4 LT |
1562 | if (retval < 0) |
1563 | goto out2; | |
1564 | ||
b58fed8b | 1565 | retval = copy_mount_options(data, &data_page); |
1da177e4 LT |
1566 | if (retval < 0) |
1567 | goto out3; | |
1568 | ||
1569 | lock_kernel(); | |
b58fed8b RP |
1570 | retval = do_mount((char *)dev_page, dir_page, (char *)type_page, |
1571 | flags, (void *)data_page); | |
1da177e4 LT |
1572 | unlock_kernel(); |
1573 | free_page(data_page); | |
1574 | ||
1575 | out3: | |
1576 | free_page(dev_page); | |
1577 | out2: | |
1578 | putname(dir_page); | |
1579 | out1: | |
1580 | free_page(type_page); | |
1581 | return retval; | |
1582 | } | |
1583 | ||
1584 | /* | |
1585 | * Replace the fs->{rootmnt,root} with {mnt,dentry}. Put the old values. | |
1586 | * It can block. Requires the big lock held. | |
1587 | */ | |
1588 | void set_fs_root(struct fs_struct *fs, struct vfsmount *mnt, | |
1589 | struct dentry *dentry) | |
1590 | { | |
1591 | struct dentry *old_root; | |
1592 | struct vfsmount *old_rootmnt; | |
1593 | write_lock(&fs->lock); | |
1594 | old_root = fs->root; | |
1595 | old_rootmnt = fs->rootmnt; | |
1596 | fs->rootmnt = mntget(mnt); | |
1597 | fs->root = dget(dentry); | |
1598 | write_unlock(&fs->lock); | |
1599 | if (old_root) { | |
1600 | dput(old_root); | |
1601 | mntput(old_rootmnt); | |
1602 | } | |
1603 | } | |
1604 | ||
1605 | /* | |
1606 | * Replace the fs->{pwdmnt,pwd} with {mnt,dentry}. Put the old values. | |
1607 | * It can block. Requires the big lock held. | |
1608 | */ | |
1609 | void set_fs_pwd(struct fs_struct *fs, struct vfsmount *mnt, | |
1610 | struct dentry *dentry) | |
1611 | { | |
1612 | struct dentry *old_pwd; | |
1613 | struct vfsmount *old_pwdmnt; | |
1614 | ||
1615 | write_lock(&fs->lock); | |
1616 | old_pwd = fs->pwd; | |
1617 | old_pwdmnt = fs->pwdmnt; | |
1618 | fs->pwdmnt = mntget(mnt); | |
1619 | fs->pwd = dget(dentry); | |
1620 | write_unlock(&fs->lock); | |
1621 | ||
1622 | if (old_pwd) { | |
1623 | dput(old_pwd); | |
1624 | mntput(old_pwdmnt); | |
1625 | } | |
1626 | } | |
1627 | ||
1628 | static void chroot_fs_refs(struct nameidata *old_nd, struct nameidata *new_nd) | |
1629 | { | |
1630 | struct task_struct *g, *p; | |
1631 | struct fs_struct *fs; | |
1632 | ||
1633 | read_lock(&tasklist_lock); | |
1634 | do_each_thread(g, p) { | |
1635 | task_lock(p); | |
1636 | fs = p->fs; | |
1637 | if (fs) { | |
1638 | atomic_inc(&fs->count); | |
1639 | task_unlock(p); | |
b58fed8b RP |
1640 | if (fs->root == old_nd->dentry |
1641 | && fs->rootmnt == old_nd->mnt) | |
1da177e4 | 1642 | set_fs_root(fs, new_nd->mnt, new_nd->dentry); |
b58fed8b RP |
1643 | if (fs->pwd == old_nd->dentry |
1644 | && fs->pwdmnt == old_nd->mnt) | |
1da177e4 LT |
1645 | set_fs_pwd(fs, new_nd->mnt, new_nd->dentry); |
1646 | put_fs_struct(fs); | |
1647 | } else | |
1648 | task_unlock(p); | |
1649 | } while_each_thread(g, p); | |
1650 | read_unlock(&tasklist_lock); | |
1651 | } | |
1652 | ||
1653 | /* | |
1654 | * pivot_root Semantics: | |
1655 | * Moves the root file system of the current process to the directory put_old, | |
1656 | * makes new_root as the new root file system of the current process, and sets | |
1657 | * root/cwd of all processes which had them on the current root to new_root. | |
1658 | * | |
1659 | * Restrictions: | |
1660 | * The new_root and put_old must be directories, and must not be on the | |
1661 | * same file system as the current process root. The put_old must be | |
1662 | * underneath new_root, i.e. adding a non-zero number of /.. to the string | |
1663 | * pointed to by put_old must yield the same directory as new_root. No other | |
1664 | * file system may be mounted on put_old. After all, new_root is a mountpoint. | |
1665 | * | |
4a0d11fa NB |
1666 | * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. |
1667 | * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives | |
1668 | * in this situation. | |
1669 | * | |
1da177e4 LT |
1670 | * Notes: |
1671 | * - we don't move root/cwd if they are not at the root (reason: if something | |
1672 | * cared enough to change them, it's probably wrong to force them elsewhere) | |
1673 | * - it's okay to pick a root that isn't the root of a file system, e.g. | |
1674 | * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, | |
1675 | * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root | |
1676 | * first. | |
1677 | */ | |
b58fed8b RP |
1678 | asmlinkage long sys_pivot_root(const char __user * new_root, |
1679 | const char __user * put_old) | |
1da177e4 LT |
1680 | { |
1681 | struct vfsmount *tmp; | |
1682 | struct nameidata new_nd, old_nd, parent_nd, root_parent, user_nd; | |
1683 | int error; | |
1684 | ||
1685 | if (!capable(CAP_SYS_ADMIN)) | |
1686 | return -EPERM; | |
1687 | ||
1688 | lock_kernel(); | |
1689 | ||
b58fed8b RP |
1690 | error = __user_walk(new_root, LOOKUP_FOLLOW | LOOKUP_DIRECTORY, |
1691 | &new_nd); | |
1da177e4 LT |
1692 | if (error) |
1693 | goto out0; | |
1694 | error = -EINVAL; | |
1695 | if (!check_mnt(new_nd.mnt)) | |
1696 | goto out1; | |
1697 | ||
b58fed8b | 1698 | error = __user_walk(put_old, LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old_nd); |
1da177e4 LT |
1699 | if (error) |
1700 | goto out1; | |
1701 | ||
1702 | error = security_sb_pivotroot(&old_nd, &new_nd); | |
1703 | if (error) { | |
1704 | path_release(&old_nd); | |
1705 | goto out1; | |
1706 | } | |
1707 | ||
1708 | read_lock(¤t->fs->lock); | |
1709 | user_nd.mnt = mntget(current->fs->rootmnt); | |
1710 | user_nd.dentry = dget(current->fs->root); | |
1711 | read_unlock(¤t->fs->lock); | |
390c6843 | 1712 | down_write(&namespace_sem); |
1b1dcc1b | 1713 | mutex_lock(&old_nd.dentry->d_inode->i_mutex); |
1da177e4 | 1714 | error = -EINVAL; |
21444403 RP |
1715 | if (IS_MNT_SHARED(old_nd.mnt) || |
1716 | IS_MNT_SHARED(new_nd.mnt->mnt_parent) || | |
1717 | IS_MNT_SHARED(user_nd.mnt->mnt_parent)) | |
1718 | goto out2; | |
1da177e4 LT |
1719 | if (!check_mnt(user_nd.mnt)) |
1720 | goto out2; | |
1721 | error = -ENOENT; | |
1722 | if (IS_DEADDIR(new_nd.dentry->d_inode)) | |
1723 | goto out2; | |
1724 | if (d_unhashed(new_nd.dentry) && !IS_ROOT(new_nd.dentry)) | |
1725 | goto out2; | |
1726 | if (d_unhashed(old_nd.dentry) && !IS_ROOT(old_nd.dentry)) | |
1727 | goto out2; | |
1728 | error = -EBUSY; | |
1729 | if (new_nd.mnt == user_nd.mnt || old_nd.mnt == user_nd.mnt) | |
1730 | goto out2; /* loop, on the same file system */ | |
1731 | error = -EINVAL; | |
1732 | if (user_nd.mnt->mnt_root != user_nd.dentry) | |
1733 | goto out2; /* not a mountpoint */ | |
0bb6fcc1 MS |
1734 | if (user_nd.mnt->mnt_parent == user_nd.mnt) |
1735 | goto out2; /* not attached */ | |
1da177e4 LT |
1736 | if (new_nd.mnt->mnt_root != new_nd.dentry) |
1737 | goto out2; /* not a mountpoint */ | |
0bb6fcc1 MS |
1738 | if (new_nd.mnt->mnt_parent == new_nd.mnt) |
1739 | goto out2; /* not attached */ | |
1da177e4 LT |
1740 | tmp = old_nd.mnt; /* make sure we can reach put_old from new_root */ |
1741 | spin_lock(&vfsmount_lock); | |
1742 | if (tmp != new_nd.mnt) { | |
1743 | for (;;) { | |
1744 | if (tmp->mnt_parent == tmp) | |
1745 | goto out3; /* already mounted on put_old */ | |
1746 | if (tmp->mnt_parent == new_nd.mnt) | |
1747 | break; | |
1748 | tmp = tmp->mnt_parent; | |
1749 | } | |
1750 | if (!is_subdir(tmp->mnt_mountpoint, new_nd.dentry)) | |
1751 | goto out3; | |
1752 | } else if (!is_subdir(old_nd.dentry, new_nd.dentry)) | |
1753 | goto out3; | |
1754 | detach_mnt(new_nd.mnt, &parent_nd); | |
1755 | detach_mnt(user_nd.mnt, &root_parent); | |
1756 | attach_mnt(user_nd.mnt, &old_nd); /* mount old root on put_old */ | |
1757 | attach_mnt(new_nd.mnt, &root_parent); /* mount new_root on / */ | |
5addc5dd | 1758 | touch_namespace(current->namespace); |
1da177e4 LT |
1759 | spin_unlock(&vfsmount_lock); |
1760 | chroot_fs_refs(&user_nd, &new_nd); | |
1761 | security_sb_post_pivotroot(&user_nd, &new_nd); | |
1762 | error = 0; | |
1763 | path_release(&root_parent); | |
1764 | path_release(&parent_nd); | |
1765 | out2: | |
1b1dcc1b | 1766 | mutex_unlock(&old_nd.dentry->d_inode->i_mutex); |
390c6843 | 1767 | up_write(&namespace_sem); |
1da177e4 LT |
1768 | path_release(&user_nd); |
1769 | path_release(&old_nd); | |
1770 | out1: | |
1771 | path_release(&new_nd); | |
1772 | out0: | |
1773 | unlock_kernel(); | |
1774 | return error; | |
1775 | out3: | |
1776 | spin_unlock(&vfsmount_lock); | |
1777 | goto out2; | |
1778 | } | |
1779 | ||
1780 | static void __init init_mount_tree(void) | |
1781 | { | |
1782 | struct vfsmount *mnt; | |
1783 | struct namespace *namespace; | |
1784 | struct task_struct *g, *p; | |
1785 | ||
1786 | mnt = do_kern_mount("rootfs", 0, "rootfs", NULL); | |
1787 | if (IS_ERR(mnt)) | |
1788 | panic("Can't create rootfs"); | |
1789 | namespace = kmalloc(sizeof(*namespace), GFP_KERNEL); | |
1790 | if (!namespace) | |
1791 | panic("Can't allocate initial namespace"); | |
1792 | atomic_set(&namespace->count, 1); | |
1793 | INIT_LIST_HEAD(&namespace->list); | |
5addc5dd AV |
1794 | init_waitqueue_head(&namespace->poll); |
1795 | namespace->event = 0; | |
1da177e4 LT |
1796 | list_add(&mnt->mnt_list, &namespace->list); |
1797 | namespace->root = mnt; | |
1798 | mnt->mnt_namespace = namespace; | |
1799 | ||
1800 | init_task.namespace = namespace; | |
1801 | read_lock(&tasklist_lock); | |
1802 | do_each_thread(g, p) { | |
1803 | get_namespace(namespace); | |
1804 | p->namespace = namespace; | |
1805 | } while_each_thread(g, p); | |
1806 | read_unlock(&tasklist_lock); | |
1807 | ||
1808 | set_fs_pwd(current->fs, namespace->root, namespace->root->mnt_root); | |
1809 | set_fs_root(current->fs, namespace->root, namespace->root->mnt_root); | |
1810 | } | |
1811 | ||
1812 | void __init mnt_init(unsigned long mempages) | |
1813 | { | |
1814 | struct list_head *d; | |
1815 | unsigned int nr_hash; | |
1816 | int i; | |
15a67dd8 | 1817 | int err; |
1da177e4 | 1818 | |
390c6843 RP |
1819 | init_rwsem(&namespace_sem); |
1820 | ||
1da177e4 | 1821 | mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount), |
b58fed8b | 1822 | 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL, NULL); |
1da177e4 | 1823 | |
b58fed8b | 1824 | mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC); |
1da177e4 LT |
1825 | |
1826 | if (!mount_hashtable) | |
1827 | panic("Failed to allocate mount hash table\n"); | |
1828 | ||
1829 | /* | |
1830 | * Find the power-of-two list-heads that can fit into the allocation.. | |
1831 | * We don't guarantee that "sizeof(struct list_head)" is necessarily | |
1832 | * a power-of-two. | |
1833 | */ | |
1834 | nr_hash = PAGE_SIZE / sizeof(struct list_head); | |
1835 | hash_bits = 0; | |
1836 | do { | |
1837 | hash_bits++; | |
1838 | } while ((nr_hash >> hash_bits) != 0); | |
1839 | hash_bits--; | |
1840 | ||
1841 | /* | |
1842 | * Re-calculate the actual number of entries and the mask | |
1843 | * from the number of bits we can fit. | |
1844 | */ | |
1845 | nr_hash = 1UL << hash_bits; | |
b58fed8b | 1846 | hash_mask = nr_hash - 1; |
1da177e4 LT |
1847 | |
1848 | printk("Mount-cache hash table entries: %d\n", nr_hash); | |
1849 | ||
1850 | /* And initialize the newly allocated array */ | |
1851 | d = mount_hashtable; | |
1852 | i = nr_hash; | |
1853 | do { | |
1854 | INIT_LIST_HEAD(d); | |
1855 | d++; | |
1856 | i--; | |
1857 | } while (i); | |
15a67dd8 RD |
1858 | err = sysfs_init(); |
1859 | if (err) | |
1860 | printk(KERN_WARNING "%s: sysfs_init error: %d\n", | |
1861 | __FUNCTION__, err); | |
1862 | err = subsystem_register(&fs_subsys); | |
1863 | if (err) | |
1864 | printk(KERN_WARNING "%s: subsystem_register error: %d\n", | |
1865 | __FUNCTION__, err); | |
1da177e4 LT |
1866 | init_rootfs(); |
1867 | init_mount_tree(); | |
1868 | } | |
1869 | ||
1870 | void __put_namespace(struct namespace *namespace) | |
1871 | { | |
1ce88cf4 | 1872 | struct vfsmount *root = namespace->root; |
70fbcdf4 | 1873 | LIST_HEAD(umount_list); |
1ce88cf4 MS |
1874 | namespace->root = NULL; |
1875 | spin_unlock(&vfsmount_lock); | |
390c6843 | 1876 | down_write(&namespace_sem); |
1da177e4 | 1877 | spin_lock(&vfsmount_lock); |
a05964f3 | 1878 | umount_tree(root, 0, &umount_list); |
1da177e4 | 1879 | spin_unlock(&vfsmount_lock); |
390c6843 | 1880 | up_write(&namespace_sem); |
70fbcdf4 | 1881 | release_mounts(&umount_list); |
1da177e4 LT |
1882 | kfree(namespace); |
1883 | } |